iHore/vLLM
8
0
Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity

[MISC] Cleanup deprecated parameters (#53559)

Browse Source 
Signed-off-by: Jee Jee Li <jeejeelee@inferact.ai>
Signed-off-by: Cyrus Leung <cyrus.tl.leung@gmail.com>
Co-authored-by: Cyrus Leung <tlleungac@connect.ust.hk>
This commit is contained in:
Jee Jee Li
2026-08-24 10:19:04 -07:00
committed by GitHub
co-authored by Cyrus Leung
parent 6648eb118d
commit 4f686e182a
12 changed files with 16 additions and 93 deletions
Download Patch File Download Diff File Expand all files Collapse all files
tests/plugins/bge_m3_sparse_plugin/bge_m3_sparse_processor/sparse_embeddings_processor.py
+4 -6
View File
@@ -58,13 +58,11 @@ class BgeM3SparseEmbeddingsProcessor(
params.dimensions = self.embed_dimensions
return params
def parse_request(
self, request_data: object
) -> SparseEmbeddingCompletionRequestMixin:
def parse_data(self, data: object) -> SparseEmbeddingCompletionRequestMixin:
# for vllm.entrypoints.llm.LLM, offline mode, calls `encode` directly.
if isinstance(request_data, dict):
return SparseEmbeddingCompletionRequestMixin(**request_data)
raise TypeError("request_data should be a dictionary")
if isinstance(data, dict):
return SparseEmbeddingCompletionRequestMixin(**data)
raise TypeError("data should be a dictionary")
def pre_process(
self,
vllm/config/attention.py
+1 -1
View File
@@ -119,7 +119,7 @@ class AttentionConfig:
if self.use_fp4_indexer_cache is not None:
logger.warning(
"use_fp4_indexer_cache is deprecated and will be removed in "
"v0.19. Use indexer_kv_dtype instead (True -> 'mxfp4')."
"v0.29. Use indexer_kv_dtype instead (True -> 'mxfp4')."
)
if self.use_fp4_indexer_cache:
if self.indexer_kv_dtype not in ("auto", "mxfp4"):
vllm/entrypoints/pooling/base/io_processor.py
+1 -1
View File
@@ -213,7 +213,7 @@ class PoolingIOProcessor:
if param.task is None:
param.task = pooling_task
elif pooling_task == "plugin":
# `plugin` task uses io_processor.parse_request to verify inputs.
# `plugin` task uses io_processor.parse_data to verify inputs.
# We actually allow plugin to overwrite pooling_task.
pass
elif param.task != pooling_task:
vllm/entrypoints/pooling/offline.py
+1 -1
View File
@@ -173,7 +173,7 @@ class PoolingOfflineMixin(OfflineInferenceMixin):
"Try converting the model using `--convert classify`."
)
# plugin task uses io_processor.parse_request to verify inputs
# plugin task uses io_processor.parse_data to verify inputs
if pooling_task != "plugin" and pooling_task != self.pooling_task:
if pooling_task not in self.supported_tasks:
raise ValueError(
vllm/entrypoints/pooling/pooling/serving.py
+1 -1
View File
@@ -74,7 +74,7 @@ class ServingPooling(PoolingBaseServing):
assert request.task is not None
pooling_task = request.task
# plugin task uses io_processor.parse_request to verify inputs
# plugin task uses io_processor.parse_data to verify inputs
if pooling_task != "plugin" and pooling_task != self.pooling_task:
if pooling_task not in self.supported_tasks:
raise VLLMValidationError(
vllm/envs.py
-19
View File
@@ -91,8 +91,6 @@ if TYPE_CHECKING:
VLLM_FLOAT32_MATMUL_PRECISION: Literal["highest", "high", "medium"] = "highest"
VLLM_BATCH_INVARIANT: bool = False
VLLM_TRITON_USE_TD: bool | None = None
# Deprecated alias of VLLM_TRITON_USE_TD (removed in v0.25).
VLLM_TRITON_ATTN_USE_TD: bool | None = None
VLLM_GPU_SYNC_CHECK: Literal["warn", "error"] | None = None
MAX_JOBS: str | None = None
NVCC_THREADS: str | None = None
@@ -563,19 +561,6 @@ def get_env_or_set_default(
logger = logging.getLogger(__name__)
def _deprecated_triton_attn_use_td() -> None:
"""Warn that VLLM_TRITON_ATTN_USE_TD was renamed to VLLM_TRITON_USE_TD.
The old name is ignored; VLLM_TRITON_USE_TD is the supported variable.
"""
if "VLLM_TRITON_ATTN_USE_TD" in os.environ:
logger.warning(
"VLLM_TRITON_ATTN_USE_TD is deprecated and will be removed in "
"v0.25. Use VLLM_TRITON_USE_TD instead."
)
return None
def _resolve_rust_cli_path() -> str | None:
"""Resolve the vllm-rs binary path.
@@ -651,10 +636,6 @@ environment_variables: dict[str, Callable[[], Any]] = {
"VLLM_GPU_SYNC_CHECK": env_with_choices(
"VLLM_GPU_SYNC_CHECK", None, ["warn", "error"]
),
# Deprecated: renamed to VLLM_TRITON_USE_TD. Kept registered so it does
# not trip the unknown-env-var check; warns on use and is otherwise
# ignored.
"VLLM_TRITON_ATTN_USE_TD": lambda: _deprecated_triton_attn_use_td(),
# Maximum number of compilation jobs to run in parallel.
# By default this is the number of CPUs
"MAX_JOBS": lambda: os.getenv("MAX_JOBS", None),
vllm/model_executor/models/mistral_large_3.py
+1 -1
View File
@@ -15,7 +15,7 @@ class MistralLarge3ForCausalLM(DeepseekV3ForCausalLM):
# (\A...\Z) and after substitution the resulting key always starts with
# "model." or "lm_head.", so no later pattern can accidentally match again.
hf_to_vllm_mapper = WeightsMapper(
orig_to_new_regex={ # noqa: B950
orig_to_new_regex={
regex.compile(
r"\Alayers\.(\d+)\.attention_norm\.weight\Z"
): r"model.layers.\1.input_layernorm.weight",
vllm/multimodal/utils.py
-14
View File
@@ -12,7 +12,6 @@ from typing import TYPE_CHECKING, Any
import numpy as np
import numpy.typing as npt
from PIL import Image
from typing_extensions import deprecated
from vllm.inputs import MultiModalPlaceholders
from vllm.utils.import_utils import LazyLoader
@@ -325,19 +324,6 @@ def group_and_batch_mm_kwargs(
yield modality, num_items, mm_kwargs_batch
@deprecated(
"`group_mm_kwargs_by_modality` has been renamed to `group_and_batch_mm_kwargs`. "
"The old name will be removed in v0.19."
)
def group_mm_kwargs_by_modality(
mm_kwargs: list[tuple[str, MultiModalKwargsItem]],
*,
device: torch.types.Device = None,
pin_memory: bool = False,
) -> Generator[tuple[str, int, BatchedTensorInputs], None, None]:
return group_and_batch_mm_kwargs(mm_kwargs, device=device, pin_memory=pin_memory)
def fetch_audio(
audio_url: str,
audio_io_kwargs: dict[str, Any] | None = None,
vllm/plugins/io_processors/interface.py
-44
View File
@@ -1,6 +1,5 @@
# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
import warnings
from abc import ABC, abstractmethod
from collections.abc import AsyncGenerator, Sequence
from typing import Generic, TypeVar
@@ -25,61 +24,18 @@ class IOProcessor(ABC, Generic[IOProcessorInput, IOProcessorOutput]):
self.vllm_config = vllm_config
def parse_data(self, data: object) -> IOProcessorInput:
if callable(parse_request := getattr(self, "parse_request", None)):
warnings.warn(
"`parse_request` has been renamed to `parse_data`. "
"Please update your IO Processor Plugin to use the new name. "
"The old name will be removed in v0.19.",
DeprecationWarning,
stacklevel=2,
)
return parse_request(data) # type: ignore
raise NotImplementedError
def merge_sampling_params(
self,
params: SamplingParams | None = None,
) -> SamplingParams:
if callable(
validate_or_generate_params := getattr(
self, "validate_or_generate_params", None
)
):
warnings.warn(
"`validate_or_generate_params` has been split into "
"`merge_sampling_params` and `merge_pooling_params`."
"Please update your IO Processor Plugin to use the new methods. "
"The old name will be removed in v0.19.",
DeprecationWarning,
stacklevel=2,
)
return validate_or_generate_params(params) # type: ignore
return params or SamplingParams()
def merge_pooling_params(
self,
params: PoolingParams | None = None,
) -> PoolingParams:
if callable(
validate_or_generate_params := getattr(
self, "validate_or_generate_params", None
)
):
warnings.warn(
"`validate_or_generate_params` has been split into "
"`merge_sampling_params` and `merge_pooling_params`."
"Please update your IO Processor Plugin to use the new methods. "
"The old name will be removed in v0.19.",
DeprecationWarning,
stacklevel=2,
)
return validate_or_generate_params(params) # type: ignore
return params or PoolingParams(task="plugin")
@abstractmethod
vllm/pooling_params.py
+1 -1
View File
@@ -88,7 +88,7 @@ class PoolingParams(
return deepcopy(self)
def verify(self, model_config: ModelConfig) -> None:
# plugin task uses io_processor.parse_request to verify inputs,
# plugin task uses io_processor.parse_data to verify inputs,
# skipping PoolingParams verify
if self.task == "plugin":
if self.skip_reading_prefix_cache is None:
vllm/v1/engine/async_llm.py
+3 -2
View File
@@ -339,8 +339,9 @@ class AsyncLLM(EngineClient):
if isinstance(prompt, EngineCoreRequest):
logger.warning_once(
"Passing EngineCoreRequest to AsyncLLM.generate() and .add_requests() "
"is deprecated and will be removed in v0.18. You should instead pass "
"the outputs of Renderer.render_cmpl() or Renderer.render_chat()."
"is deprecated and will be removed in the future. You should "
"instead pass the outputs of Renderer.render_cmpl() or "
"Renderer.render_chat()."
)
request = prompt
vllm/v1/engine/llm_engine.py
+3 -2
View File
@@ -236,8 +236,9 @@ class LLMEngine:
if isinstance(prompt, EngineCoreRequest):
logger.warning_once(
"Passing EngineCoreRequest to LLMEngine.generate() and .add_requests() "
"is deprecated and will be removed in v0.18. You should instead pass "
"the outputs of Renderer.render_cmpl() or Renderer.render_chat()."
"is deprecated and will be removed in the future. You should "
"instead pass the outputs of Renderer.render_cmpl() or "
"Renderer.render_chat()."
)
request = prompt